Upload folder using huggingface_hub

README.md

@@ -67,7 +67,8 @@ uv run python scripts/hydrate_defextra.py \
 
 - [`docs/defextra_hydration.md`](docs/defextra_hydration.md) (technical details, CLI flags, markers).
 - [`docs/get_pdfs.md`](docs/get_pdfs.md) (how to find PDFs).
-- [`docs/mismatch_examples.md`](docs/mismatch_examples.md) (real examples of mismatch types and fixes).
+- [`docs/mismatch_examples.md`](docs/mismatch_examples.md) (mismatch types with short excerpts).
+- [`docs/defextra_legal_diff.md`](docs/defextra_legal_diff.md) (source vs HF legal CSV diff).
 - See [`docs/defextra_hydration.md`](docs/defextra_hydration.md) for technical details and [`docs/get_pdfs.md`](docs/get_pdfs.md) for PDF sources.
 
 ## Expected minor mismatches

docs/defextra_hydration.md

@@ -128,4 +128,4 @@ uv run python scripts/hydrate_defextra.py \
 - Small mismatches are expected due to PDF/GROBID text normalization.
 - Missing exact TEI spans do **not** block hydration; hash/anchor markers are used as fallback.
 - Exact TEI spans are validated against stored hashes; if they do not match, citation‑stripped hash/anchor matching is used instead.
-- See [`docs/mismatch_examples.md`](mismatch_examples.md) for concrete examples and fixes.
+- See [`docs/mismatch_examples.md`](mismatch_examples.md) for concrete mismatch types with short excerpts.

docs/mismatch_examples.md

@@ -1,90 +1,68 @@
 # Mismatch classes (hydration vs reference)
 
-This page summarizes common difference types seen in the latest hydrated run.
-Examples are **short fragments** with surrounding text removed.
-
-- Total differences: 144
-- wording_change: 74
-- punctuation_only: 50
-- casing: 8
-- hyphenation: 5
-- digit_letter_spacing: 3
-- truncation: 2
-- citation_spacing: 1
-- header_or_boilerplate: 1
-
----
-## Wording / lexical differences
+This page shows **short excerpts** around observed differences between the hydrated output
+and the reference dataset (`defextra_hf_tablefix.csv`).
+Excerpts are intentionally tiny and only highlight the mismatch itself.
+
+Observed classes (examples shown below):
 
-Small wording changes (e.g., singular/plural) or tokenization artifacts.
+- wording: 3 example(s)
+- punctuation: 3 example(s)
+- casing: 3 example(s)
+- hyphenation: 3 example(s)
+- digit_spacing: 3 example(s)
+- truncation: 2 example(s)
+- header: 1 example(s)
+
+## Wording / lexical differences
 
 | Paper | Concept | Field | Reference (fragment) | Hydrated (fragment) |
 | --- | --- | --- | --- | --- |
-| 016f7a076ac272db106fbcea056752c7307f676a | specification error | context | …Conducted mainly in the late 1970s and 1980s, wave 3 of researchwitnessed yet another round o… | …Conducted mainly in the late 1970 s and 1980 s, wave 3 of research witnessed yet another round… |
-| 0209e602acaeab882fee84e244caf574cf345ef9 | ratio bias/numerosity bias | context | …ault. In the classic ratio bias task derived from Piaget and Inhelder (1951/1975), participants are offered a prize if th… | …ault. In the classic ratio bias task derived from Piaget andInhelder (1951/1975), participants are offered a prize if th… |
-| 033b21cf1c6d3bdae587e673452b994443bf3546 | narrative | context | …he bare minimum Simply put, narrative isthe representation of a… | …bare minimum Simply put, narrative is the representation of … |
+| 057bd5a3a6b652dfd1419128aae578478f5891b0 | stereotypes | definition | Negative beliefs and attitudes assigned … | negative beliefs and attitudes assigned … |
+| 09c660d438e8b372f3e77504c069255b85c00256 | centrality | definition | …nging a high-level feature has a greaterimpact on the meaning of an object than … | …nging a high-level feature has a greater impact on the meaning of an object than… |
+| 09c660d438e8b372f3e77504c069255b85c00256 | psychological distance | definition | a subjective experience that something i… | expect a relatively high sensitivity to … |
 
 ## Punctuation-only differences
 
-Only punctuation differs (e.g., comma, quote style).
-
 | Paper | Concept | Field | Reference (fragment) | Hydrated (fragment) |
 | --- | --- | --- | --- | --- |
-| 0f7eda998bbce003745ff2fdbcaa1d9a8119368b | echo chamber | context | …ces of news, then we impose on ourselves a narrowed and selfreinforcing epistemic filter, which leaves out contrary view… | …ces of news, then we impose on ourselves a narrowed and self-reinforcing epistemic filter, which leaves out contrary view… |
-| 17734113f254a64b3bae312713edba3b1e34fb56 | Post-truth Era | context | …Indeed, today we live in what some have called a “post-truth” era, which is characterized by digital disinform… | …Indeed, today we live in what some have called a "post-truth" era, which is characterized by digital disinform… |
-| 1901.00596v4 | network embedding | definition | …ional vector representation of a node which preserves a node’s topological information… | …ional vector representation of a node which preserves a node's topological information… |
+| 033b21cf1c6d3bdae587e673452b994443bf3546 | narrative | definition | The representation of an event or a seri… | the representation of an event or a seri… |
+| 03c4bb2121b3dd2b5302d37c81e84092a501bc91 | platformed racism | definition | …ordances, business models and policies-- and the specific cultures of use associ… | …ordances, business models and policies--and the specific cultures of use associa… |
+| 2106.16117v1 | long-term energy security | definition | …mic developments and environmental needs | …mic developments and environmental needs' |
 
 ## Casing differences
 
-Same text except for upper/lower case.
-
 | Paper | Concept | Field | Reference (fragment) | Hydrated (fragment) |
 | --- | --- | --- | --- | --- |
-| 033b21cf1c6d3bdae587e673452b994443bf3546 | narrative | definition | …The representation of an event or a series of events.… | …the representation of an event or a series of events.… |
-| 05bfced33d92944b7a0672490c371342d28ee076 | observational bias | definition | …Observed data differs systematically from the unobserved data… | …observed data differs systematically from the unobserved data… |
-| 1538c4777271ae6abb542801dac01423f4d566ad | publication bias | definition | …Significant results are more likely to be published while non… | …significant results are more likely to be published while non… |
+| 05bfced33d92944b7a0672490c371342d28ee076 | observational bias | definition | Observed data differs systematically fro… | observed data differs systematically fro… |
+| 1538c4777271ae6abb542801dac01423f4d566ad | publication bias | definition | Significant results are more likely to b… | significant results are more likely to b… |
+| 24589ef86242ff182e7bae495b94e7c11931f20c | anxious semiotics | definition | Sense of uncertainty and concern over po… | sense of uncertainty and concern over po… |
 
 ## Hyphenation / line-break joins
 
-Hyphenation caused by line breaks is joined differently.
-
 | Paper | Concept | Field | Reference (fragment) | Hydrated (fragment) |
 | --- | --- | --- | --- | --- |
-| 3584741 | computational narrative representations | context | …cused review, we will focus exclusively on event-based narra- tive representations. Thus, we define computational narrativ… | …cused review, we will focus exclusively on event-based narrative representations. Thus, we define computational narrativ… |
-| 3584741 | metro maps method | context | …ro maps [98, 99] are an extension of the Connect the Dots ap- proach that represents more than a single storyline using a … | …ro maps [98, 99] are an extension of the Connect the Dots approach that represents more than a single storyline using a … |
-| 3584741 | open source intelligence (OSINT) | context | …]. Although OSINT data sources leverage more than just tradi- tional news articles [38], OSINT could still benefit from ne… | …]. Although OSINT data sources leverage more than just traditional news articles [38], OSINT could still benefit from ne… |
+| 3457607 | AI Fairness 360 | definition | …kit developed by IBM to help moving fair- ness research algorithms into an indus… | …kit developed by IBM to help moving fairness research algorithms into an industr… |
+| 1901.00596v4 | Directed Graph | context | …rected from one node to another. An undi- rected graph is considered as a specia… | …rected from one node to another. An undirected graph is considered as a special … |
+| 1901.00596v4 | GNNs | context | …graph-related tasks in an end-to-end man- ner. Many GNNs explicitly extract high… | …graph-related tasks in an end-to-end manner. Many GNNs explicitly extract high-l… |
 
 ## Letter–digit spacing
 
-Spacing between digits/letters differs (`bias4` vs `bias 4`).
-
 | Paper | Concept | Field | Reference (fragment) | Hydrated (fragment) |
 | --- | --- | --- | --- | --- |
-| 0090023afc66cd2741568599057f4e82b566137c | omitted variable bias | context | …Omitted Variable Bias. Omitted variable bias4 occurs when one or more important variables are left out o… | …Omitted Variable Bias. Omitted variable bias 4 occurs when one or more important variables are left out o… |
-| 016f7a076ac272db106fbcea056752c7307f676a | selection bias | context | …cts of race on sentencing. Conducted mainly in the late 1970s and 1980s, wave 3 of research witnessed yet another round … | …cts of race on sentencing. Conducted mainly in the late 1970 s and 1980 s, wave 3 of research witnessed yet another round… |
-| 235c4f33d5bfc81bfa09a2458fcc0e42ef4454dc | propaganda | context | …It published workbooks and held seminars in the early 1930s aimed at promoting the ideal of "self-determination," rega… | …It published workbooks and held seminars in the early 1930 s aimed at promoting the ideal of "self-determination," rega… |
+| 0090023afc66cd2741568599057f4e82b566137c | omitted variable bias | context | …ted Variable Bias. Omitted variable bias4 occurs when one or more important vari… | …ted Variable Bias. Omitted variable bias 4 occurs when one or more important var… |
+| 016f7a076ac272db106fbcea056752c7307f676a | selection bias | context | …ncing. Conducted mainly in the late 1970s and 1980s, wave 3 of research witnesse… | …ncing. Conducted mainly in the late 1970 s and 1980 s, wave 3 of research witnes… |
+| 1901.00596v4 | VC dimension | context | …the VC dimension of a GNN* [15] is O(p 4n 2 ) if it uses the sigmoid or tangent … | …the VC dimension of a GNN* [15] is O(p 4 n 2) if it uses the sigmoid or tangent … |
 
 ## Truncation (missing tail)
 
-Hydrated text is missing the end of the reference span.
-
 | Paper | Concept | Field | Reference (fragment) | Hydrated (fragment) |
 | --- | --- | --- | --- | --- |
-| 016f7a076ac272db106fbcea056752c7307f676a | specification error | definition | …the omission of explanatory variable… | …the omission of explanatory variables… |
-| 0d23df558a30492946059c017343a431dc3dc172 | inter-media agenda-setting | definition | …ed theory to explain how content transfers between news medi… | …ed theory to explain how content transfers between news media… |
-
-## Citation spacing/formatting differences
-
-Citation formatting differs (e.g., `[155, 164]` vs `[155,164]`).
-
-| Paper | Concept | Field | Reference (fragment) | Hydrated (fragment) |
-| --- | --- | --- | --- | --- |
-| frai-06-1225093 | Explainability | context | …ndent, and AI embraces a wide variety of tasks (Miller, 2019a). We treat explainability as a specialization of interpret… | …ndent, and AI embraces a wide variety of tasks (Miller, 2019 a). We treat explainability as a specialization of interpret… |
-
-## Header/boilerplate inserted
+| 016f7a076ac272db106fbcea056752c7307f676a | specification error | definition | the omission of explanatory variable | the omission of explanatory variables |
+| 0d23df558a30492946059c017343a431dc3dc172 | inter-media agenda-setting | definition | … how content transfers between news medi | … how content transfers between news media |
 
-Hydrated text includes header/boilerplate not in reference.
+## Header / boilerplate inserted
 
 | Paper | Concept | Field | Reference (fragment) | Hydrated (fragment) |
 | --- | --- | --- | --- | --- |
-| https://arxiv.org/abs/2312.16148 | spin bias | context | …ndencies between words and phrases must be considered [110]. Spin Bias describes a form of bias introduced either by lea… | …ndencies between words and phrases must be considered [110]. Manuscript submitted to ACM The Media Bias Taxonomy Spin Bias describes a form of bias introduced either by lea… |
+| https://arxiv.org/abs/2312.16148 | spin bias | context | …ds and phrases must be considered [110]. Spin Bias describes a form of bias intr… | …ds and phrases must be considered [110]. Manuscript submitted to ACM The Media B… |

pyproject.toml

@@ -2,14 +2,14 @@
 name = "defextra-dataset"
 version = "0.1.0"
 description = "DefExtra hydration utilities (legal release)"
-requires-python = ">=3.10"
+requires-python = ">=3.12,<3.13"
 dependencies = [
-  "grobid-client-python>=0.0.9",
-  "lxml>=4.9",
-  "pdfplumber>=0.10",
-  "pdfminer.six>=20221105",
-  "PyPDF2>=3.0",
-  "pyarrow>=14.0",
+  "grobid-client-python>=0.1.2",
+  "lxml>=6.0.2",
+  "pdfplumber>=0.11.9",
+  "pdfminer.six>=20251230",
+  "PyPDF2>=3.0.1",
+  "pyarrow>=22.0.0",
 ]
 
 [tool.uv]

scripts/hydrate_defextra.py

@@ -1047,36 +1047,6 @@ def main() -> None:
             )
             ctx_strip_brackets, ctx_strip_parens = _strip_flags(row, "context")
 
-            if not definition and pdf_token_index:
-                span = _find_pdf_hash_span(row, pdf_token_index, "definition")
-                if span:
-                    definition = _extract_with_trailing_punct(
-                        pdf_token_index.doc_text,
-                        span[0],
-                        span[1],
-                    )
-                    hydrated_from_pdf += 1
-                if not definition:
-                    stripped_index = _get_stripped_index(
-                        pdf_token_cache_stripped,
-                        pdf_path,
-                        pdf_token_index.doc_text,
-                        def_strip_brackets,
-                        def_strip_parens,
-                    )
-                    if stripped_index is not None:
-                        span = _find_pdf_hash_span(
-                            row,
-                            stripped_index,
-                            "definition",
-                        )
-                        if span:
-                            definition = _extract_with_trailing_punct(
-                                stripped_index.doc_text,
-                                span[0],
-                                span[1],
-                            )
-                            hydrated_from_pdf += 1
             if not definition and tei_token_index:
                 spec = _select_hash_specs(row, "definition")
                 if spec:
@@ -1239,32 +1209,31 @@ def main() -> None:
                 )
                 if _span_matches_hash(row, candidate, "definition"):
                     definition = candidate
-
-            if not context and pdf_token_index:
-                span = _find_pdf_hash_span(row, pdf_token_index, "context")
+            if not definition and pdf_token_index:
+                span = _find_pdf_hash_span(row, pdf_token_index, "definition")
                 if span:
-                    context = _extract_with_trailing_punct(
+                    definition = _extract_with_trailing_punct(
                         pdf_token_index.doc_text,
                         span[0],
                         span[1],
                     )
                     hydrated_from_pdf += 1
-                if not context:
+                if not definition:
                     stripped_index = _get_stripped_index(
                         pdf_token_cache_stripped,
                         pdf_path,
                         pdf_token_index.doc_text,
-                        ctx_strip_brackets,
-                        ctx_strip_parens,
+                        def_strip_brackets,
+                        def_strip_parens,
                     )
                     if stripped_index is not None:
                         span = _find_pdf_hash_span(
                             row,
                             stripped_index,
-                            "context",
+                            "definition",
                         )
                         if span:
-                            context = _extract_with_trailing_punct(
+                            definition = _extract_with_trailing_punct(
                                 stripped_index.doc_text,
                                 span[0],
                                 span[1],
@@ -1433,6 +1402,36 @@ def main() -> None:
                 )
                 if _span_matches_hash(row, candidate, "context"):
                     context = candidate
+            if not context and pdf_token_index:
+                span = _find_pdf_hash_span(row, pdf_token_index, "context")
+                if span:
+                    context = _extract_with_trailing_punct(
+                        pdf_token_index.doc_text,
+                        span[0],
+                        span[1],
+                    )
+                    hydrated_from_pdf += 1
+                if not context:
+                    stripped_index = _get_stripped_index(
+                        pdf_token_cache_stripped,
+                        pdf_path,
+                        pdf_token_index.doc_text,
+                        ctx_strip_brackets,
+                        ctx_strip_parens,
+                    )
+                    if stripped_index is not None:
+                        span = _find_pdf_hash_span(
+                            row,
+                            stripped_index,
+                            "context",
+                        )
+                        if span:
+                            context = _extract_with_trailing_punct(
+                                stripped_index.doc_text,
+                                span[0],
+                                span[1],
+                            )
+                            hydrated_from_pdf += 1
 
         if not definition and pdf_path is not None and pdf_token_index:
             spec = _select_hash_specs(row, "definition")